MICA · FIRST-LINE LEARNING NOTES
Understand, prioritise and organise follow-up
Appraise a diagnostic or therapeutic development before changing first-line practice.
Identify what is genuinely new
A new name, a recent publication or a claimed performance does not establish benefit. Restate the question: better diagnosis, improved clinical outcomes, lower risk or a simpler pathway? Identify the document type, comparator and population. Guidance, a trial and a technology demonstration provide different types of information.
[1][2]Read a trial beyond the headline
For a therapeutic intervention, examine eligibility, group allocation, the primary outcome, follow-up duration, missing data and adverse effects. CONSORT 2025 supports transparent reporting of this information; completeness alone does not prove freedom from bias. Check whether the result concerns a patient-important event or only an intermediate marker.
[1]For a new development, distinguish technical performance, clinical benefit and implementation. Statistical significance may accompany a small absolute benefit or a surrogate outcome. Check follow-up, excluded populations and harms. For AI, good test-set performance does not establish improved clinical decisions or safety in another setting.
[1][2]Communicate benefit accurately
Use the same time horizon and denominator for benefits and harms. Accompany relative change with absolute context and uncertainty. Distinguish the average observed effect from expected benefit for this individual. Discuss burden, monitoring, alternatives and the option of not immediately changing care when medically appropriate.
[3]- Read the primary outcome and absolute results rather than only headlines, relative percentages or press releases.
- Before using AI, look for external validation, a relevant comparator and consequences of errors.
- Maintain human oversight and verify sources; generated suggestions are not validated recommendations.
Maintain clinical oversight of AI tools
Generative models may produce plausible but false, biased or incomplete answers. WHO highlights automation bias among the risks. Compare a proposal with available clinical information and original sources. Confidentiality and clinical responsibility remain essential; an educational demonstration does not validate a tool for every patient or use.
[2]A learning situation
FICTIONAL CASE · SELF-REFLECTION WITHOUT A SCORE
In a fictional example, an intervention reduces an event from 2 people in 100 to 1 in 100 over the same period. A headline mentions only a 50% reduction.
How should the finding be explained clearly?
Read the reasoning
Present both frequencies and the absolute reduction of one event per 100 people over that period, then uncertainty and adverse effects. A 50% relative reduction is consistent with these numbers but does not on its own describe the size of benefit. This arithmetic example is not a real trial.
References and scope
- Hopewell S et al. · CONSORT 2025 statement: Updated guideline for reporting randomised trials · PLOS Medicine ↗
- OMS · Orientations sur les grands modèles multimodaux · communiqué du 18 janvier 2024 ↗
- NICE NG197 · Shared decision making · recommandations ↗
Original educational synthesis for adult first-line practice. International references must be interpreted with the clinical setting and local care pathways. Independent human scientific review remains to be confirmed.